shib

• http://github.com/tagomoris/shib
• Blog entry tagomoris.tumblr

DESCRIPTION

Shib is web client application for SQL-like query engines, written in Node.js, supporting

• Hive (hiveserver, hiveserver2)
• Facebook Presto
• Google BigQuery

Once configured, we can switch query engines per executions.

Some extension features are supported:

• Setup queries: options to specify queries executed before main query, like 'create functions ...'
• Default Database: option to specify default database for Hive 0.6 or later

Versions

Latest version of 'shib' is v1.0.2.

'shib' versions are:

• v1.0 series
  • metadata of v1.0 is NOT compatible with v0.3, so migration required
• v0.3 series
  • use latest node (~> v0.10.26)
  • multi engines/databases support
  • presto support
  • storages of v0.3.x are compatible with v0.2
  • engine-wide database/table access controls
  • tagging for executed queries
  • authentication / logging for query execution
  • access controls based on authentication
• v0.2 series
  • current status of master branch
  • uses local filesystem instead of KT, depends on latest node (v0.8.x, v0.10.x)
  • higher performance, more safe Web UI and updated features
  • storages of v0.2 are NOT complatible with v0.1
• v0.1 series
  • uses KT, depends on node v0.6.x
  • see v0.1 tag

INSTALL

Hive/Presto

For Hive queries, shib requires HiveServer or HiveServer2. Setup and run these.

• For HiveServer2
  • Configure hive.server2.authentication as NOSASL
    • Strongly recommended to configure hive.support.concurrency as false

For Presto, shib is tested with Presto version 0.57.

Node.js

To run shib, you must install node.js (v0.10.x recommended), and export PATH for installed node.

shib

Clone shib code.

$ git clone git://github.com/tagomoris/shib.git

Install libraries, configure addresses of HiveServer (and other specifications).

$ npm install
$ vi config.js

And run.

$ npm start

Shib listens on port 3000. see http://localhost:3000/

To switch environments for each shib instance, use NODE_ENV environment variable. (ex: production.js will be used with NODE_ENV=production)

$ NODE_ENV=production NODE_PATH=lib node app.js

Migrate metadata database from v0 to v1

Migration operation required to execute shib v1, with data in v0 era.

1. Stop shib process
2. Update shib code to v1
3. Execute npm run migrate (for var/database.sqlite3 file)

• This operation requires 5 minutes 30 seconds for 220MB database
• Backup v0 database file is var/database.sqlite3.v0

4. Start shib

Configuration

Shib can have 2 or more query executor engines.

HiveServer2

Basic configuration with HiveServer2 in config.js (or productions.js):

var servers = exports.servers = {
  listen: 3000,
  fetch_lines: 1000,   // lines per fetch in shib
  query_timeout: null, // shib waits queries forever
  setup_queries: [],
  storage: {
    datadir: './var'
  },
  engines: [
    { label: 'mycluster1',
      executer: {
        name: 'hiveserver2',
        host: 'hs2.mycluster1.local',
        port: 10000,
        username: 'hive',
        support_database: true
      },
      monitor: null
    },
  ],
};

username should be same as user name that hive job will be executed on. (password is not required for NOSASL transport.)

For UDFs, you can specify statements before query executions in setup_queries.

var servers = exports.servers = {
  listen: 3000,
  fetch_lines: 1000,
  query_timeout: null,
  setup_queries: [
    "add jar /path/to/jarfile/foo.jar",
    "create temporary function foofunc as 'package.of.udf.FooFunc'",
    "create temporary function barfunc as 'package.of.udf.BarFunc'"
  ],
  storage: {
    datadir: './var'
  },
  engines: [
    { label: 'mycluster1',
      executer: {
        name: 'hiveserver2',
        host: 'hs2.mycluster1.local',
        port: 10000,
        support_database: true
      },
      monitor: null
    },
  ],
};

HiveServer

Classic HiveServer is available if you want database supports instead of HiveServer2.

var servers = exports.servers = {
  listen: 3000,
  fetch_lines: 1000,
  query_timeout: null,
  setup_queries: [],
  storage: {
    datadir: './var'
  },
  engines: [
    { label: 'mycluster1',
      executer: {
        name: 'hiveserver',  // HiveServer(1)
        host: 'hs1.mycluster1.local',
        port: 10000,
        support_database: true,
        default_database: 'mylogs1'
      },
      monitor: null
    },
  ],
};

Presto

For Presto, use presto executer.

var servers = exports.servers = {
  listen: 3000,
  fetch_lines: 1000,
  query_timeout: 30, // 30 seconds for Presto query timeouts (it will fail)
  setup_queries: [],
  storage: {
    datadir: './var'
  },
  engines: [
    { label: 'prestocluster1',
      executer: {
        name: 'presto',
        host: 'coordinator.mycluster2.local',
        port: 8080,
        user: 'shib',
        catalog: 'hive',  // required configuration argument
        support_database: true,
        default_database: 'mylogs1'
      },
      monitor: null
    },
  ],
};

BigQuery

For BigQuery, use bigquery executer.

var servers = exports.servers = {
  listen: 3000,
  fetch_lines: 1000,
  query_timeout: 30, // 30 seconds for BigQuery query timeouts (it will fail)
  storage: {
    datadir: './var'
  },
  engines: [
    { label: 'bigquery',
      executer: {
        name: 'bigquery',
        default_database: 'mylogs1',
        project_id: 'gcp-project-id',
        key_filename: '/path/to/keyfile.json'
      },
      monitor: null
    }
  ]
};

For more detail about project_id and key_filename config, see https://github.com/GoogleCloudPlatform/gcloud-node#authorization .

Multi clusters and engines

Shib supports 2 or more engines for a cluster, and 2 or more clusters with same engines. These patterns are available.

• HiveServer1, HiveServer2 and Presto for same data source
• 2 or more catalogs for same Presto cluster
• Many clusters which has one of HiveServer, HiveServer2 or Presto

You should write configurations in engines how you wants. fetch_lines, query_timeout and setup_queries in each engines overwrites global default of these configurations.

For example, This is configuration example.

• ClusterA has HiveServer2
  • listenes port 10000
  • customized udfs in foo.jar are availabe
• ClusterB has HiveServer
  • listenes port 10001
  • customized udfs in foo.jar are available
• Presto cluster is configured with hive catalog and hive2 catalog
  • udfs are not available

var servers = exports.servers = {
  listen: 3000,
  fetch_lines: 1000,
  query_timeout: null,
  setup_queries: [
    "add jar /path/to/jarfile/foo.jar",
    "create temporary function foofunc as 'package.of.udf.FooFunc'",
    "create temporary function barfunc as 'package.of.udf.BarFunc'"
  ],
  storage: {
    datadir: './var'
  },
  engines: [
    { label: 'myclusterA',
      executer: {
        name: 'hiveserver2',
        host: 'master.a.cluster.local',
        port: 10000,
        support_database: true
      },
      monitor: null
    },
    { label: 'myclusterB',
      executer: {
        name: 'hiveserver',
        host: 'master.b.cluster.local',
        port: 10001,
        support_database: true,
        default_database: 'mylogs1'
      },
      monitor: null
    },
    { label: 'prestocluster1',
      executer: {
        name: 'presto',
        host: 'coordinator.p.cluster.local',
        port: 8080,
        user: 'shib',
        catalog: 'hive',
        support_database: true,
        default_database: 'mylogs1',
        query_timeout: 30,  // overwrite global config
        setup_queries: []   // overwrite global config
      },
      monitor: null
    },
    { label: 'prestocluster2',
      executer: {
        name: 'presto',
        host: 'coordinator.p.cluster.local',
        port: 8080,
        user: 'shib',
        catalog: 'hive2',  // one engine config per catalogs
        support_database: true,
        default_database: 'default',
        query_timeout: 30,  // overwrite global config
        setup_queries: []   // overwrite global config
      },
      monitor: null
    }
  ],
};

Access Control

Shib have access control list for Databases/Tables. Default is 'allow' for all databases/tables.

Shib's access control rules are:

• configure per executer
• database level default ('allow' or 'deny') without any optional rules makes that database unvisible
• database level default + allow/deny table list makes its tables visible/unvisible
  • in this case, this 'database' is visible
• default 'allow' or 'deny' decides visibilities of databases without any optional rules

'unvisible' databases and tables:

• are not shown in tables/partitions list and schema list
• cannot be queried by users (these queries always fails)

Access control options are written in 'executer' like this:

executer: {
  name: 'presto',
  host: 'coordinator.p.cluster.local',
  port: 8080,
  catalog: 'hive',
  support_database: true,
  default_database: 'default',
  query_timeout: 30,
  setup_queries: [],
  access_control: {
    databases: {
      secret: { default: "deny" },
      member: { default: "deny", allow: ["users"] },
      test:   { default: "allow", deny: ["secretData", "userMaster"] },
    },
    default: "allow"
  }
},

For more details, see wiki: Access Control.

Monitors

monitor configurations are used to get query status and to kill queries.

JobTracker (MRv1)

jobtracker monitor is available in MRv1 environment (w/ both of hiveserver and hiveserver2).

monitor: {
  name: 'jobtracker',
  host: 'jobtracker.hostname.local',
  port: 50030,
  mapred: '/usr/bin/mapred' // 'mapred' in PATH by default
}

For this feature, shib should be executed by a user who can execute command mapred job -kill JOB_ID.

YARN (MRv2)

yarn monitor is available in MRv2 environment (w/ both of hiveserver and hiveserver2).

monitor: {
  name: 'yarn',
  host: 'resourcemanager.hostname.local',
  port: 8088
}

In this case, shib kills query with Resource Manager REST API.

If you specify yarn command description, shib kills query with yarn application -kill APP_ID.

monitor: {
  name: 'yarn',
  host: 'resourcemanager.hostname.local',
  port: 8088,
  yarn: '/usr/bin/yarn'
}

Huahin Manager (obsolete)

For monitors in CDH4 + MRv1 environment, Huahin manager is available.

To show map/reduce status, and/or to kill actual map/reduce jobs behind hive query, shib can use Huahin Manager. Current version supports only 'Huahin Manager CDH4 + MRv1' only.

http://huahinframework.org/huahin-manager/

Configure monitor argument like below instead of null.

monitor: {
  name : 'huahin_mrv1',
  host: 'localhost',
  port: 9010
}

Authentication

Shib have authentication to log who execute queries and to control accesses: setup_queries_auth option means to specify queries executed before main query when authentication is required.

var servers = exports.servers = {
  listen: 3000,
  fetch_lines: 1000,   // lines per fetch in shib
  query_timeout: null, // shib waits queries forever
  setup_queries: [],
  setup_queries_auth: ["set hive.mapred.mode=strict"],
  storage: {
    datadir: './var'
  },
  auth: {
    type: 'http_basic_auth',
    url: 'http://your.internal.protected.service.example.com/',
    realm: '@your.service.example.com'
  },
  engines: [
    { label: 'mycluster1',
      executer: {
        name: 'hiveserver2',
        host: 'hs2.mycluster1.local',
        port: 10000,
        username: 'hive',
        support_database: true
      },
      monitor: null
    },
  ],
};

For more details, see wiki: Authentication.

Miscellaneous configurations

Disable "history" tab

Specify disable_history: true on servers.

var servers = exports.servers = {
  listen: 3000,
  fetch_lines: 1000,
  query_timeout: null, // seconds. (null:shib will wait query response infinitely).
  setup_queries: [],
  disable_history: true,
  storage: {
    datadir: './var'
  },

As HTTP Proxy for query engines

POST query string into /execute with some parameters.

curl -s -X POST -F 'querystring=SELECT COUNT(*) AS cnt FROM yourtable WHERE field="value"' http://shib.server.local:3000/execute | jq .
{
  "queryid": "69927e67c5b1d5f665697943cc4867ec",
  "results": [],
  "dbname": "default",
  "engine": "hiveserver",
  "querystring": "SELECT COUNT(*) AS cnt FROM yourtable WHERE field=\"value\""
}

Specify engineLabel and dbname for non-default query engines and databases:

curl -s -X POST -F "engineLabel=presto" -F "dbname=testing" -F "querystring=SELECT COUNT(*) AS cnt FROM yourtable WHERE field='value'" http://shib.server.local:3000/execute

If you want not to add your query into history tab, specify 'scheduled':

curl -s -X POST -F "scheduled=true" -F "querystring=SELECT COUNT(*) AS cnt FROM yourtable WHERE field='value'" http://shib.server.local:3000/execute

Then, fetch query's status whenever you want.

curl -s http://shib.server.local:3000/status/69927e67c5b1d5f665697943cc4867ec 
executed

Or get whole query object.

curl -s http://shib.server.local:3000/query/69927e67c5b1d5f665697943cc4867ec | jq .
{
  "queryid": "69927e67c5b1d5f665697943cc4867ec",
  "results": [
    {
      "resultid": "969629614dff69411a2f4f1733c9616a",
      "executed_at": "Wed Feb 26 2014 16:02:00 GMT+0900 (JST)"
    }
  ],
  "dbname": "default",
  "engine": "hiveserver",
  "querystring": "SELECT COUNT(*) AS cnt FROM yourtable WHERE field=\"value\""
}

If this query object has executed status, or a member of results, you can fetch its result by resultid.

# if you want elasped time or bytes or lines or ....
curl -s http://shib.server.local:3000/result/969629614dff69411a2f4f1733c9616a | jq .
{
  "schema": [
    {
      "type": "bigint",
      "name": "cnt"
    }
  ],
  "completed_msec": 1393398893759,
  "completed_at": "Wed Feb 26 2014 16:14:53 GMT+0900 (JST)",
  "completed_time": null,
  "bytes": 6,
  "queryid": "69927e67c5b1d5f665697943cc4867ec",
  "executed_time": null,
  "executed_at": "Wed Feb 26 2014 16:14:52 GMT+0900 (JST)",
  "executed_msec": 1393398892752,
  "resultid": "969629614dff69411a2f4f1733c9616a",
  "state": "done",
  "error": "",
  "lines": 2
}
# raw result data as TSV (fast)
curl -s http://shib.server.local:3000/download/tsv/969629614dff69411a2f4f1733c9616a | jq .
CNT
1234567
# or CSV (slow)
curl -s http://shib.server.local:3000/download/csv/969629614dff69411a2f4f1733c9616a | jq .
"CNT"
"1234567"

These HTTP requests/response are same with that javascript on browser does.

---

TODO

• Paches are welcome!

License

Copyright 2011- TAGOMORI Satoshi (tagomoris)

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.